Chinese Named Entity Recognition via Multi-Channel Attention of Lexicon

Lexicon, as a kind of external knowledge, has been widely used by existing studies to assist the model for identifying entity boundaries effectively. However, most existing approaches only pay attention to the words related to a certain entity and do not consider the impact of words of different lengths on the recognized entity. In this paper, we propose a neural model for named entity recognition tasks enhanced by integrating multi-channel attention. In the multi-channel attention module, we assign words to different channels according to their length and measure the degree of attention the words have to the entity. Experiments results on three widely used Chinese benchmark datasets for NER demonstrate the effectiveness of our method.